Optical communication systems, particularly those associated with telecommunications and data center applications, face an ever-increasing need for larger optical switching configurations, such as optical cross-connects and “fiber-to-free-space” switching fabrics. The interconnectivity of devices via Internet-based cloud computing, as well as cloud storage capabilities, has raised the demand for lower cost optical communication systems that are able to easily and quickly switch signal paths.
Optical routing of one path of an N×N array of input ports to any other output port of an N×N array of outputs will be further enabled by the development of a readily producible two-dimensional (2D) fiber array component that is available for a relatively low cost. In such free-spacing routing paradigms, an N×N fiber array is disposed at the back focal plane of an N×N lens array, which produces beams with their minimum waist between the communicating arrays. The routing function is enabled, for example, by a pair of properly-placed, two-dimensional MEMS arrays that allows any input to be switched to any output. FIG. 1 illustrates an exemplary structure for providing this function. Note that the optical arrangement is typically a conjugate imaging system so that a position error of the fiber at the input will be beam position error at the output fiber. Very small changes (i.e., on the order of a micron) create large insertion loss errors. This type of error is often referred to as a positional error related to variations in the center-to-center spacing between the core regions of adjacent optical fibers in the array (hereinafter referred to as “pitch”).
Besides the inter-fiber spacing (pitch) inaccuracies, error in the pointing of the beam as it exits a fiber will produce a displacement of the output beam that can create coupling loss, clipping, and scattered noise problems. FIG. 2 illustrates this “pointing error” for an exemplary optical fiber 1, as positioned through an aperture 3 formed in a substrate 4. As shown, optical fiber 1 passes through aperture 3 in an off-axis manner, creating an angular amount of fiber tilt (θ), measured with respect to a normal of an exit surface 2. This measure θ is defined as the angular pointing error. While shown in this case as maintaining its optical axis, insertion of optical fibers through apertures may also result in the fiber experience some amount of bending, also creating angular pointing error at the output. Fiber and thus bean pointing error impinging onto a collimating lens produces an output beam that is displaced relative to the optical axis of the system. To accommodate this displacement (so as to avoid clipping, scattering, and cross-talk), the MEMS micro-mirrors would need to increase in size, with the undesirable effect of reducing their density or increasing the complexity of the MEMS design. Thus, problems associated with creating free-space optical cross-connections, in high volume resides with the 2D fiber array requiring a low pitch error, as well as a low pointing error. Today's applications for such a switching fabric have simultaneous requirements of a pitch error on the order of ±1 μm or less, and a pointing error on the order of ±15 mrad (or smaller).
To date, one approach to improve the 2D fiber array component is based upon the utilization of a multiple number of precisely-etched (tapered) silicon wafers, each wafer formed to include progressively smaller and more accurately aligned vias, which may require high hole aspect ratios (i.e., the ratio of the side wall straight length to the hole diameter). The cost of fabricating multiple silicon wafers with different-sized vias, and then manipulating a stack of these wafers to align the vias is prohibitive from a cost point of view (although the required, precise alignment may be achieved). Furthermore, it is more costly and difficult to produce high aspect ratios as previously described. Locating individual wafers farther from each other axially can help address pointing error issues, but increases the difficulty of assembly of such 2D arrays, and ultimately increases the cost.
In another approach, only a pair of wafers is used, where their vias are aligned and then fibers are inserted one at a time (or one 1×N fiber array at a time) and positioned to create the desired alignment. Here, the assembly time is significant and cumbersome, again resulting in an expensive process. Additionally, since each element of this configuration is a precisely made component, the final structure can be costly.
U.S. Pat. Nos. 6,470,123 and 6,766,086 are illustrative of these prior art techniques. U.S. Pat. No. 6,470,123, which issued to Sherman et al. on Oct. 22, 2002, describes a high density optical fiber array assembly and assembly method that utilizes a series of separate, stacked guide plates that form a series of fiber guide channels. The guide plates are stacked within a housing so that the bottom of one acts as a cover for the channels of another. The fiber arrays can be “tool inserted” along the channels as one group, such as a row of fibers, or manually inserted one at a time and advanced sequentially. U.S. Pat. No. 6,766,086, which issued to Sherman et al. on Jul. 20, 2004 describes an optical fiber array apparatus comprising a housing front mask having a matrix of fiber seating openings, with each opening having one or more side walls. An optical fiber extends through each opening and a tool is used to press the fiber side surface into engagement with the one or more side walls to precisely position and secure the fiber. Bonding material then fills all voids in and around the opening. In one embodiment, a clamping wafer behind the front mask moves to clamp the fibers to the front mask opening walls. In another, the front mask defines flexing arms with distal ends that clamp fibers to opening walls and in yet another elongated flexible members lie along front mask slots to clamp fibers in openings that communicate into the slots.